Interpreting pronouns and demonstratives in Finnish ...emkaiser/kaiser-trueswell2008-LCP-w.pdf · Interpreting pronouns and demonstratives in Finnish: Evidence for a form-speciﬁc

Interpreting pronouns and demonstratives in Finnish:

Evidence for a form-specific approach to reference

resolution

Elsi KaiserDepartment of Linguistics, University of Southern California, Los Angeles,

CA, USA

John C. TrueswellDepartment of Psychology, University of Pennsylvania, Philadelphia,

PA, USA

Two Finnish language comprehension experiments are presented which suggestthat the referential properties of pronouns and demonstratives cannot bereduced straightforwardly to the salience level of the antecedent. The findings,from a sentence completion study and visual-world eye-tracking study, revealan asymmetry in which features of the antecedent Finnish pronouns anddemonstratives are most sensitive to, both in terms of their final interpretations

Correspondence should be addressed to Elsi Kaiser, Department of Linguistics, 3601 Watt

Way, GFS 301, University of Southern California, Los Angeles, CA 90089-1693, USA. E-mail:

[email protected]

This article is based on research reported in Elsi Kaiser’s dissertation (2003, University of

Pennsylvania) and presented at the 16th Annual CUNY Conference on Human Sentence

Processing in Boston in March 2003, and at the Architectures and Mechanisms for Language

Processing conference in Glasgow in August 2003. Preliminary analyses of some of the data

reported here appeared in Kaiser and Trueswell (in press).

Thanks to Jennifer Arnold, Cassie Creswell, Eleni Miltsakaki, Kimiko Nakanishi, Ritva

Laury, Ellen Prince, Jeffrey Runner, Maribel Romero, Michael Tanenhaus, Jennifer Venditti, and

the psycholinguistics lab groups at the University of Pennsylvania and the University of

Rochester for many useful comments and suggestions. Many thanks also to all those in Finland

who made these experiments possible. We gratefully acknowledge Kelly Rulison, Mike

Muscianesi, and Noreen Stackhouse for their help with stimulus preparation and coding.

This research was partially supported by a grant from the National Institutes of Health

(1-R01-HD37507) awarded to the second author.

LANGUAGE AND COGNITIVE PROCESSES

2008, 23 (5), 709�748

# 2008 Psychology Press, an imprint of the Taylor & Francis Group, an Informa business

http://www.psypress.com/lcp DOI: 10.1080/01690960701771220

Downloaded By: [USC University of Southern California] At: 06:11 28 April 2010

and during real-time processing. In particular, the syntactic role and linearposition of the antecedent, two factors which have been claimed to influencereferent salience, have different effects on the interpretation of pronouns anddemonstratives. Contrary to what is commonly assumed, pronouns anddemonstratives cannot be mapped onto a unified salience hierarchy, becausethey exhibit different degrees of sensitivity to syntactic role and word order. Weoffer an alternative approach to anaphor resolution, the form-specific multiple-constraints approach.

INTRODUCTION

During language comprehension, we are continuously occupied with the task

of reference resolution. To understand a sentence, a comprehender needs to

add to her mental model the new information it conveys. To do this

successfully, she must also link it to what she already knows. Pronouns (e.g.,

‘they’, ‘s/he’) and demonstratives (e.g., ‘this’, ‘that’) play a central role in this

information management process and are frequent in both written and

spoken language. However, due to their shorthand nature, they appear to

pose a challenge for successful comprehension. Pronouns and demonstra-

tives are informationally impoverished and do not, on their own, provide

enough information to identify the intended referent.

It is generally agreed that there exists a correlation between the type of

referential form used to refer to an entity and the level of salience/

prominence of the entity: The more reduced an anaphoric expression

is, the more prominent its referent needs to be in the speaker’s and listener’s

mental models of the discourse (e.g., Arnold, 1998, p. 4). This view is

reflected in the referential form hierarchies that have been proposed by a

range of researchers, including Gundel, Hedberg, and Zacharski (1993),

Givon (1983) and Ariel (1990). Part of the standard hierarchy is shown

in (1), where the forms that are further to the left are used for more salient

referents.

(1) null � pronoun � demonstrative � full Noun Phrase . . .

Connecting reduced referential forms and salient referents seems plau-

sible, since ‘[a]n expression that has little semantic content . . . can contribute

little or nothing to the identification process, and can only be used where

identification of the referent is either straightforward or not an issue’

(Garnham, 2001, p. 55). However, referential forms cannot always be

distinguished based on their informativeness. For example, the English

pronoun it and the demonstratives this/that ‘are indistinguishable with

respect to the description they provide for the intended referent (an

inanimate object)’ (Ariel, 2001, p. 29). Nevertheless, it is commonly agreed

710 KAISER AND TRUESWELL


that demonstratives are used for less salient referents than pronouns.1 Many

salience hierarchy-based approaches specifically claim that pronouns have

more salient antecedents than demonstratives, and that the referential

properties of different forms follow from their positions on the hierarchy,

not from differences in informativeness.In this paper, we take a closer look at the referential properties of

pronouns and demonstratives in Finnish, in order to investigate the validity

of the salience hierarchy approach and to improve our understanding of

what it means for a referent to be salient/prominent. Based on the results of a

sentence completion study and an eye-tracking experiment, we suggest that

the referential properties of pronouns and demonstratives cannot be reduced

straightforwardly to the salience level of the antecedent. Our data reveal an

asymmetry in terms of which features of the antecedent Finnish pronouns

and demonstratives are most sensitive to, both in terms of their final

interpretations and during real-time processing. In particular, our findings

show that the syntactic role and linear position of the antecedent, two factors

which have been claimed to influence referent salience, have different effects

on the interpretation of pronouns and demonstratives. In light of these

findings, we argue that pronouns and demonstratives cannot be mapped

onto a unified salience hierarchy, because they differ in terms of the

properties they ‘value’ most highly in their antecedents.

Which factors influence referent salience?

Previous research has found that reference resolution � assumed to be an

indicator of referent salience � is influenced by a range of factors, including

syntactic role, word order, information structure (distinctions such as old/

given vs. new information), anaphoric form, discourse connectives and verb

semantics (see Arnold, 1998 for an overview).

Effects of syntactic role on referent salience have been observed many

times. Researchers claim that, at least in English, (agentive) subjects are more

salient than non-subjects (e.g., Chafe, 1976; Brennan, Friedman, & Pollard,

1987; Crawley & Stevenson, 1990; see also Gordon, Grosz, & Gilliom, 1993,

Gordon & Chan, 1995. For research on non-agentive subjects see Turan,

1995, 1998; Di Eugenio, 1998; Gordon & Chan, 1995).

What grants subjects this special prominence? In a language like English,

with relatively fixed subject-object order, there are at least two possible

1 The fact that some referential forms in some languages provide information about things

such as number, gender, animacy, or ‘humanness’ (e.g., English ‘it’ vs. ‘she/he’) etc. is, in our

opinion, incontrovertible and thus we do not address it here (see Arnold et al., 2000; Greene,

McKoon, & Ratcliff, 1992; Garrod & Sanford, 1982; Albrecht & Clifton, 1998; inter alia, for

work on gender and number). In this paper we focus on choices in referential form that cannot

be explained by these kinds of factors.

PRONOUNS AND DEMONSTRATIVES IN FINNISH 711


explanations for this: (i) subjects precede objects and (ii) agentive subjects

differ thematically/semantically from objects. To disentangle these possibi-

lities, researchers have turned to languages with flexible word order. The

results, so far, are rather controversial. For German, Rambow (1993) claims

that word order can influence reference resolution (see Gernsbacher &

Hargreaves, 1988; Gordon et al., 1993 on order-of-mention in English).

Strube and Hahn (1996, 1999) make similar claims for German, arguing that

hearer-old (familiar/known to the hearer) entities are ranked above hearer-

new ones (see Prince, 1992). It follows from these claims that if information

structural factors (especially hearer-status) guide word order, word order

influences referent salience (see also Hajicova & Vrbova, 1982; Hajicova,

Kubon, & Kubon, 1990). In contrast, Turan (1998) and Hoffman (1998)

claim that in Turkish, referent salience correlates with grammatical (or

semantic) role, not word order. Prasad and Strube (2000) make similar claims

for Hindi. According to these approaches, the factors driving word order

variation are independent of reference resolution (see also Vallduvı, 1993).

These conflicting findings may stem partly from the fact that word order

variation does not occur for the same reasons in all languages and all

constructions. A constituent can be located in a noncanonical position for a

range of reasons, e.g., because it has been previously mentioned, or because

it contrasts with something else in the discourse (e.g., Kiss, 1995; Vilkuna,

1995). In this paper, we focus on the effects of word order variation driven by

a particular kind of discourse information, namely the given/new distinction.

One factor or multiple factors?

So far in this discussion, it has been implicitly assumed that a single factor

determines salience. However, in contrast to the views of Hoffman and

Strube and Hahn, other researchers, including Ariel (1990), Arnold (1998),

and Lappin and Leass (1994), regard salience as a ‘compound’ notion

resulting from the interaction of multiple factors or constraints. Various

psycholinguistic experiments (e.g., Arnold et al., 2000; Badecker & Straub,

2002; Gordon et al., 1993; Jarvikivi et al., 2005) also suggest that anaphor

resolution can be influenced by multiple constraints.

Thus, loosely speaking, there are two main approaches to referent

salience: (i) the single-factor approach, which assumes that one factor � be

it word order, linear order or something else � determines a referent’s

salience, and (ii) the multiple-factor approach, which assumes that multiple

weighted constraints contribute towards a referent’s salience. In this paper,

based on earlier work by Kaiser (2003, 2005b), we explore an idea that

extends the second approach, the multiple constraints view (see also Brown-

Schmidt, Byron, & Tanenhaus, 2004; Brown-Schmidt, Byron, & Tanenhaus,

2005). We hypothesise that not only is reference resolution sensitive to



multiple constraints, but that different anaphoric forms � at least those that

are informationally equivalent � can differ in how sensitive they are to

different factors. According to this form-specific approach, it could be the

case that one particular anaphoric form is primarily sensitive to word order,

and another form is primarily sensitive to syntactic role, or sensitive to both

word order and syntactic role but to different degrees. (We use the term

‘(anaphoric) form’ to refer to different referential expressions, i.e., different

lexical items. Pronouns, demonstratives and full nouns are treated as

different anaphoric forms. Different inflected forms of a particular pronoun

(e.g., s/he inflected for different cases) count as occurrences of the same

anaphoric form.) What is at stake is the question of whether or not

anaphoric forms align along a unified salience scale. Even though the single-

factor approach and the multiple-factor approach differ in their view of what

contributes to a referent’s salience level, to the best of our understanding

they both agree that different anaphoric forms can be characterised in terms

of the salience of their antecedents; for example, pronouns refer to more

salient referents than demonstratives. The form-specific multiple-constraints

approach, in contrast, allows for a situation where the referential properties

of anaphoric forms cannot be described in terms of a unified notion of

antecedent salience.

Finnish

Two main typological characteristics of Finnish make it well-suited for

investigating whether anaphoric forms differ in their sensitivities to different

information types. First, Finnish displays discourse-driven word order

variation, which allows us to uncouple contributions of word order and

syntactic role. Second, standard Finnish has two kinds of third person

anaphors than can be used to refer to humans, namely the gender-neutral

pronoun han ‘s/he’ and the demonstrative tama ‘this’. Given the claims of

accessibility-hierarchy-based approaches, these forms can be used as tools to

test the salience of referents.Finnish word order is very flexible. The canonical order is SVO, but all six

permutations of S, V, and O are grammatical in the appropriate contexts

(Vilkuna, 1989, 1995). We focus on the information-structural properties of

SVO vs. OVS variation. The choice between SVO and OVS is guided by

whether or not the arguments have been mentioned in the preceding

discourse (e.g., Chesterman, 1991; Hiirikoski, 1995; see also Helasvuo,

2001 on pronominal subjects). Since standard Finnish has no definite or

indefinite article, the SVO/OVS variation plays an important role in

conveying the distinction between discourse-new and discourse-old informa-

tion. (For discussion of other word orders, including the role of contrastive

focus in SOV and OSV order, see Vilkuna, 1989, 1995.)



Subjects in a noncanonical, postverbal position are used to refer to

discourse-new referents, as in (2a). In contrast, subjects in a canonical

preverbal position usually refer to entities mentioned in preceding discourse,

as in (2b). In a discourse-initial all-new sentence, a preverbal subject can also

be interpreted as referring to a discourse-new entity.2

(2a) Pylvaaseen nojasi solakka tummahiuksinen nainen (Remes, 1997, p. 369)

column-ILL leaned slim-NOM dark-haired-NOM woman-NOM)

‘A slim, dark-haired woman leaned against the column.’

(2b) Nainen puhui hanen kanssaan saksaa. (Remes, 1997, p. 343))

woman-NOM spoke he-GEN with German-PART)

‘The woman spoke German with him.’

An object in a noncanonical preverbal position in an OVS sentence, as in

(3a), is interpreted as discourse-old information. In the canonical postverbal

position, objects can be interpreted as new or old information, as in (3b) and

(3c). Valimaa-Blum (1988) notes that, when no contrastive focus is present,

sentences with preverbal subjects and postverbal subjects can both have the

default intonation pattern of Finnish, i.e., a gradually descending F0 contour

(see also Iivonen, 1998; Sulkala & Karjalainen, 1992). The sentences

discussed in this paper do not involve contrastive focus.

(3a) autoa ajoi kolmas henkilo

car-PART drove third-NOM person-NOM

‘a third party drove the car.’

(from the newspaper Viitasaaren Seutu, June 2006)

(3b) kunnes eras meista osti auton

until one-NOM us-ELAT bought car-ACC

‘until one of us bought a car’

(from www.ttk.oulu.fi/tyoharjoittelu/raportit/Leverkusen_TMD4.htm)

(3c) Aalto osti auton

Aalto-NOM bought car-ACC

‘(a man called) Aalto bought the car’ [context: discussing the fate of a 1965

Ford]

(from the newspaper Aamulehti 4/7/2006)

The anaphoric paradigm of Finnish has two forms that can be used to

refer to third person human referents: the gender-neutral pronoun han ‘s/he’

and the demonstrative tama ‘this’. The pronoun han is commonly described

as referring to the most important character in a particular situation, to the

character in the foreground (e.g., Kalliokoski, 1991; see also Vilppula, 1989).

2 Abbreviations used: NOM�nominative case, ACC�accusative, PART�partitive, GEN�genitive, ILL�illative, ELAT�elative, INESS�inessive, ADESS�adessive, poss�possessive

suffix.



According to Saarimaa (1949), subjects are more in the foreground than

other constituents and thus han tends to refer to entities in subject position.

Evidence supporting this claim comes from a corpus study by Halmari

(1994) who found that han is usually used to refer back to a preceding

subject.3 An example is given in (4).

(4a) Koskela marssitti joukkueensa parakin eteen.

Koskela-NOM marched troops-ACC-poss barrack-GEN in-front-of

‘Koskela had his troops march to the front of the barracks.’

(4b) Kotvan han seisoskeli aivan kuin miettien miten aloittaisi. (Linna, 1954/1999,

p. 10)

moment-ACC he-NOM stood exactly as-if thinking how

start-CONDITIONAL

‘He stood there for a moment, as if thinking about how he should start.’

(4c) # Kotvan tama seisoskeli aivan kuin miettien miten aloittaisi.

The demonstrative tama ‘this’ has multiple uses in Finnish. Similar to the

English proximal demonstrative ‘this’, it can be used as a demonstrative

pronoun or a discourse deictic (e.g., Etelamaki, 1996). Consider, for example,

‘This is my sister’ and ‘James pushed Julie. This surprised her’. However,

unlike English ‘this,’ tama can also be used anaphorically to refer to third

person human referents. Our focus in this paper is on this particular use of

tama. In contrast to han, tama is often characterised as referring to

characters in the background (e.g., Varteva, 1998). According to Sulkala

and Karjalainen (1992), tama is ‘used to indicate the last mentioned out of

two or more possible referents’ (1992, pp. 282�283). An example is given in

(5). Crucially, we cannot replace han in (4b) with tama (4c), because use of

tama in these kinds of contexts hinges on the presence of a second salient

referent, and using tama to refer to the only salient referent results in

infelicity.4

(5a) Koskela alkoi tuijottaa laulavaa vanrikkia.

Koskela-NOM started to-stare singing-PART second-lieutenant-PART

‘Koskela started to stare at the singing second lieutenant.’

(5b) Tama jatkoi aluksi lauluaan, mutta alkoi sitten vaivautua . . . (Linna, 1954/

1999, p. 285)

3 In this paper, we focus on standard Finnish. Dialects of colloquial Finnish have somewhat

different anaphoric systems, see e.g., Laitinen (1992), Seppanen (1998) and Vilppula (1989).4 As discussed in Kaiser (in press), there are certain special contexts, involving logophoricity

(i.e., from the perspective of the person whose speech, thoughts, or feelings are being reported),

where tama can be used to refer to what seems, at first blush, to be a salient referent. However, as

Kaiser (in press) shows, a closer look at these kinds of examples shows that even in these

contexts (where the logophoric referent is very salient), the defining characteristics of tama is

that it is used to refer to characters that are not the most salient, not at the centre of attention.



This-NOM continued in-beginning song-poss, but started then to-be-

bothered

‘First he (second lieutenant) continued singing, but then started to be a bit

bothered’

In sum, since Finnish has flexible word order and two anaphoric forms

that can be used for third person human antecedents, it provides a good

testing ground for evaluating the validity of the form-specific multiple-

constraints approach.

The question of whether han and tama are influenced by word order in the

same way has not received much attention in the Finnish anaphor resolution

literature. In one of the earliest papers on han and tama, Saarimaa (1949)

appears to support a single-factor approach relying on grammatical role, as

his claims suggest that tama refers to a recent non-subject, and han to a

subject. A more recent study by Halmari (1994), based on intuitions from

seven native speakers, suggests that with SVO order, pronouns prefer subjects

and demonstratives prefer objects. However, Halmari did not obtain clear

results for OVS order, perhaps due to absence of context and the presence of

plausibility biases in her materials. Existing corpus studies on han and tama

in standard Finnish (e.g., Halmari, 1994; Kaiser, 2000) have failed to shed

much light on this matter, due to the difficulties of finding sufficiently large

numbers of the relevant types of examples in a non-tagged corpus.

Some preliminary evidence in favour of the form-specific multiple-

constraints approach is provided by the results of a small-scale sentence

completion experiment on the interpretation of han and tama following SVO

and OVS order, reported in Kaiser (2003, 2005b). The results suggested that

han and tama are affected in different ways by word order. Both SVO/Han

and OVS/Han conditions exhibited a subject preference regardless of word

order; the pronoun han was interpreted as referring to the subject in 63% and

61% of the cases respectively. (On the interpretation of han following SVO/

OVS order, see also Jarvikivi, van Gompel, Hyona, and Bertram, 2005,

discussed in the General Discussion section.) In contrast, with tama,

changing the word order from SVO to OVS had a striking effect. In the

SVO/Tama condition, tama showed a strong object preference (83% object

continuations), but in the OVS/Tama condition, tama was split between

subject and object (33% and 38% respectively, as well as some other types of

continuations. See also discussion of Experiment 1 below). Thus, the results

of this sentence completion experiment (Kaiser 2003, 2005b) show that, in

contrast to han’s sensitivity to syntactic role, with tama both linear order and

syntactic role matter.However, this experiment used only a small number of target items and

presented the SVO and OVS sentences out of context. The absence of context

introduces a serious asymmetry between the SVO and the OVS conditions,



given that the SVO/OVS variation is driven by the discourse status of the

subject and object. An SVO sentence with no preceding context is perceived

as more felicitous, and processed significantly faster and more easily, than an

OVS sentence out of context (Halmari, 1994; Hyona & Hujanen, 1997).

Adding a supportive discourse context significantly decreases the processing

difficulties associated with OVS order (see Kaiser & Trueswell, 2004). Thus,

the absence of preceding context introduces a strong felicity asymmetry

between SVO and OVS. Given that OVS order is crucial to disentangling the

contributions of syntactic role and word order, testing it in a situation where

it is known to be infelicitous and hard to process raises concerns regarding

the validity of the findings.

Aims of this paper

Our main aim is to investigate the validity of the single-factor and multiple-

factor approaches to reference resolution � in particular the feasibility of the

form-specific approach � by testing how different kinds of information

influence the referential properties of Finnish pronouns and demonstratives.

We first looked at the off-line referential properties of han and tama by means

of a sentence-completion experiment, using context to ensure that both SVO

and OVS sentences are felicitous. Second, we investigated the real-time

processing of these forms by means of the visual-world eye-tracking

paradigm, to gain insights into the time-course with which different factors

influence reference resolution. With an on-line measure such as eye-tracking

we can find out whether effects of syntactic role and word order are

temporally distinct or whether they both influence processing simultaneously.

General predictions

The word order single-factor approach predicts that word order, and in

particular the information structure it encodes, determines the referential

properties of han and tama, presumably starting at the earliest stages of

processing. This predicts that when han/tama is preceded by a sentence with

SVO or OVS order, han is interpreted as referring to the preverbal and tama

to the postverbal argument. In contrast, the syntactic role single-factor

approach claims that from the onset of reference resolution processes, han is

interpreted as referring to an entity whose most recent antecedent was in

subject position5 and tama to an entity whose most recent antecedent was in

object position, regardless of word order.

5 In the rest of this paper, we will often say, for the sake of brevity, that the pronoun/

demonstrative refers to the subject/object � even though this is not strictly speaking correct, since

the referential form actually, in the end, picks out the entity whose most recent antecedent was in

subject/object position.



The multiple constraints view offers a third possibility: both word order

and syntactic role have an effect. The prediction is that han and tama are

both sensitive to some combination of word order and syntactic role,6

potentially even during the earliest stages of processing. For SVO order, thepredictions are straightforward, since word order and syntactic role are

aligned: han is predicted to refer to the subject, and tama to the object. For

OVS order, the precise predictions depend on the relative weights of word

order and syntactic role. If word order and syntactic role are weighted

equally, neither han nor tama will show a clear preference for subject or

object; both forms should be split between the two potential referents.

Regardless of the weights, this approach predicts that han and tama will

divide up the referential labour evenly, such that they will be interpreted asreferring to distinct referents.

The form-specific approach claims that multiple factors play a role but

also allows for the possibility that han and tama do not exhibit the same

degree of sensitivity to word order and syntactic role. The results of the

preliminary sentence completion study (Kaiser, 2005b) suggest that, at least

in terms of final interpretation, this is indeed the case. On the basis of that

experiment, we predict that (i) the pronoun han is sensitive primarily to

syntactic role and prefers subjects regardless of word order, and (ii) thedemonstrative tama is sensitive to a combination of syntactic role and word

order, such that in SVO order it prefers the postverbal object, and in OVS

order it is split between the postverbal subject and the preverbal object.

EXPERIMENT 1: SENTENCE COMPLETION

Method

Participants. Sixteen adult native Finnish speakers volunteered forparticipation in the experiment. They were not paid for their participation.

They were recruited over the internet.

Materials. This experiment used a standard sentence completion task to

investigate the effects of word order and syntactic role on the referential

properties of han and tama. Each target item consisted of a short narrative

that contained an SVO or OVS sentence, which was followed by the first

word of the next sentence (in nominative case, which signals that it is thesubject), either han ‘s/he’ or tama ‘this’. Anaphor type and word order were

crossed to create four conditions: (a) SVO/Han, (b) OVS/Han, (c) SVO/

Tama, (d) OVS/Tama. The critical SVO/OVS sentences were preceded by a

6 In this paper we focus on the effects of word order and syntactic role. For related research

on the effects of anaphoric form, see Kaiser (2003), Kaiser (2005a).



brief, two-sentence context which mentions the entity that the preverbal

argument of the critical sentence refers to (i.e., S in SVO, O in OVS). The

postverbal argument of the critical sentence is not mentioned in the

preceding context. Thus, both SVO and OVS sentences are felicitous,

because in both orders the preverbal argument is discourse-old and the

post-verbal argument is discourse-new.7 A sample item in the SVO/Han

condition is given in (6).

(6) Niina oli ostoksilla ruokakaupassa.

Niina-NOM was shopping-ADESS grocery-store-INESS

‘Niina was shopping at the grocery store.’ [first sentence]

Jonossa odottaessaan han naki takanaan valkohattuisen kokin.

Line-INESS waiting-INESS-poss she-NOM saw behind-poss white-hatted-

ACC cook-ACC

‘While waiting in line, she saw a cook with a white hat behind her.’

[second sentence]

Kokki toni jonon hannilla seisovaa leipuria.

Cook-NOM pushed line-GEN tails-ADESS standing-PART baker-PART

‘The cook-SUBJ pushed a baker-OBJ standing at the back of the line.’

[SVO critical sentence]

or Kokkia toni jonon hannilla seisova leipuri.

Cook-PARTpushedline-GENtails-ADESSstanding-NOMbaker-NOM

‘A baker-SUBJ standing at the back of the line pushed the cook-OBJ.’

[OVS critical sentence]

Han . . .

S/he-NOM. . .

‘S/he . . .’ [prompt word]

In order to be able to felicitously refer to both referents in the critical SVO/

OVS sentence (the third sentence) with full NPs, another referent was also

introduced into the context (e.g., Niina in (6)). The first sentence introduces

this referent, and at the beginning of the second sentence, s/he is referred to

with a pronoun.

7 Word order (first vs. second mention) was confounded with repeated mention; the first

mentioned character had already been mentioned, using a full noun. Existing work indicates that

repeated mention can contribute to reference resolution and/or referent topicality in English (see

e.g. Kameyama, 1996; Levy, 1982; see also Ariel, 1990). For simplicity, we refer to effects of this

factor as a word-order effect. However, we recognise that any such effects could also be due to

repeated mention. The conclusions we draw at the end of this paper do not hinge upon this

interpretation choice.



There were 16 critical items and 32 fillers. Four presentation lists were

constructed by combining the 16 target stories with the 32 filler stories. The

target items were separated from each other by at least one filler item. Within

a presentation list, eight of the target trials appeared with the SVO structure

and eight appeared with the OVS structure. For each of these sentence

structure types, four were followed by the pronoun han and four by the

demonstrative tama. Each target item was then rotated through these four

conditions, generating four different presentation lists. Reverse order lists

were also generated to control for trial order. The nouns used for the subject

and object in the critical items were all professions or other ‘roles’ (e.g.,

doctor, stewardess, reporter, student), and all verbs had agentive subjects.

Procedure and data analysis

Participants were asked to provide natural-sounding continuations for the

sentence fragments. Participation took place over the internet, via a web-

page where participants could type in their responses. Participants’

continuations were coded according to which of the referents in the

preceding sentence (the first- or second-mentioned NP) the participants

chose as the referent of the pronoun. When it was not clear which referent

the participant had interpreted as being the referent of the pronoun or

demonstrative, the continuation was marked as ‘unclear’. In addition, if the

demonstrative tama was not used as an anaphor for one of the characters

mentioned in the preceding sentence (e.g., it was used as a discourse-deictic,

as in ‘This was rather strange’), the continuation was coded as a

‘demonstrative’ use. Examples are in Appendix 1.

Results

Table 1 shows the percentage of different continuation types for each of the

four conditions. Continuations were grouped based on whether the

TABLE 1Percentages of different continuation types in the four conditions of the sentence

continuation task, showing which referent the anaphoric expression was used to referto. (The number in parentheses is the actual number of continuations, out of a total of

64 continuations in each of the four conditions.)

1st-mentioned ref 2nd-mentioned ref Dem Unclear/other

SVO.Han 64% (41) 13% (8) 0% (0) 23% (15)

OVS.Han 13% (8) 64% (41) 0% (0) 23% (15)

SVO.Tama 0% (0) 88% (56) 9% (6) 3% (2)

OVS.Tama 9% (6) 44% (28) 30% (19) 17% (11)



participant interpreted the pronoun/demonstrative as referring to the first-

mentioned referent of the previous sentence (subject in S¯VO, object in O

¯VS),

the second-mentioned referent (object in SVO¯, subject in OVS

¯), or other. In

addition, the data for first-mentioned and second-mentioned continuations

are plotted in Figure 1. As the figure illustrates, the pronoun han shows a

subject preference regardless of word order. In both the SVO/Han condition

and the OVS/Han condition, han is interpreted as referring to the subject of

the preceding sentence in 64% of the continuations. This resembles the results

of the out-of-context experiment (Kaiser, 2005b), which found 63% subject

continuations in the SVO/Han condition and 61% subject continuations in

the OVS/Han condition.

The results for the demonstrative tama in the SVO and OVS conditions

are less parallel. With SVO order, tama prefers the second-mentioned (i.e.,

postverbal) object (88%), which mirrors the results of the earlier out-of-

context experiment. With OVS order, tama has a preference for the

postverbal subject over the preverbal object (44% subject continuations,

9% object continuations) � but this is a weaker preference than in the SVO/

Tama condition. It also differs from the out-of-context experiment, where

tama preceded by OVS order was split between subject and the object (33%

subject interpretations, 38% object interpretations).

Table 1 also shows the number of so-called ‘demonstrative’ continuations,

i.e., cases where people treated tama not as an anaphor but as a

demonstrative or a discourse deictic. The number of demonstrative inter-

pretations is higher in the OVS/Tama condition than in the SVO/Tama

condition (30% and 9% respectively). A possible explanation for this is

discussed below.

To analyse the data statistically, we used first-mentioned advantage scores.

These scores were calculated by subtracting the proportion of continuations

Figure 1. Percentage of continuations in which the anaphoric expression is interpreted as

referring to the first-mentioned referent or the second-mentioned referent in the sentence

continuation task (Experiment 1), plotted for each of the four conditions.



referring to the second-mentioned referent from the proportion of continua-

tions referring to the first-mentioned referent. A positive number indicates

more continuations referring to the first-mentioned referent than to the

second-mentioned one, and a negative number signals that there were more

continuations referring to the second-mentioned referent than the first-

mentioned one.

The mean first-mentioned advantage scores were calculated for each

participant in each condition and each item in each condition. The resulting

participant and item means were entered into separate analyses of variance

(ANOVAs). All ANOVAs had these four factors: Word order (SVO or OVS),

Anaphor type (pronoun or demonstrative), Order (forward or reverse list)

and List (four levels) in the participant analysis and Item Group (four

groups) in the item analysis.

The ANOVAs show that the first-mentioned advantage scores (i.e., how

strongly an anaphoric element prefers the first-mentioned referent over the

second-mentioned post-verbal referent) are significantly influenced by

anaphor type: han or tama, F1(1, 8)�178.94, pB.001, F2(1, 12)�51.27,

pB.001, and word order (SVO or OVS, F1(1, 8)�11.38, p�.01, F2(1, 12)�22.59, pB.001.8 There is also a significant interaction, F1(1, 8)�64.93,

pB.001, F2(1, 12)�49.34, pB.001.

In other words, whether an anaphoric expression is interpreted as

referring to the preceding subject or object depends on whether the anaphor

is han or tama, and whether the word order of the preceding sentence is SVO

or OVS. In addition, the significant interaction between word order and

anaphor type reveals the asymmetrical behaviour of han and tama. As Figure

1 shows, word order has no effect on han, but does have an effect on tama.

Discussion

The results of Experiment 1 show that word order has strikingly different

effects on the interpretation pronouns and demonstratives. The pronoun han

is sensitive primarily to syntactic role, and prefers subjects regardless of word

order. In contrast, the demonstrative tama prefers postverbal referents (i.e.,

O in SVO and S in OVS), but this preference is modulated by the syntactic

role of the postverbal argument: tama prefers objects over subjects.

As a whole, the results of Experiment 1 do not fit with the single-factor

approaches, nor with a multiple-constraint approach assuming all forms to

be subject equally to the same constraints. One of the single-factor

predictions outlined above was that grammatical role determines what han

8 There were reliable effects involving control variables in some of the analyses we conducted

in this paper. Because we believe that they have no bearing on the proposals that we will be

making, these effects will not be reported.



and tama refer to, with han predicted to refer to the subject, and tama to the

object, regardless of order. This prediction does indeed seem to match the

results for han.

If syntactic role is crucial for locating the antecedent of the pronounhan, then it seems natural to expect that it will also guide the interpretation

of the demonstrative tama. However, this is not the case: tama shows a

significant preference for the postverbal referent with both SVO and OVS

order. At first glance, the results for tama appear to be compatible with the

prediction that only word order is relevant. Nevertheless, despite showing a

preference for postverbal referents, tama is also sensitive to syntactic role:

There are more postverbal object continuations in the SVO/Tama

continuation than there are postverbal subject continuations in the OVS/Tama condition, indicating that effects of linear order are modulated by

syntactic role. Our claim that tama’s sensitivity to word order is modulated

by syntactic role receives further support from the distribution of

demonstrative continuations in the different conditions, as shown in Table

1: OVS/Tama prompts more demonstrative continuations than SVO/Tama.

Using tama demonstratively rather than anaphorically can be regarded as

signalling that neither the subject nor the object is a good enough

antecedent for tama � in other words, it provides participants with an‘escape hatch’ in a situation where neither argument is a good antecedent

for tama. Thus, the increased proportion of demonstrative uses in OVS/

Tama (30%) as compared to SVO/Tama (9%) suggests that postverbal

subjects are not as well-suited to be antecedents of tama as postverbal

objects.

Interestingly, in the original experiment that did not provide a felicitous

context for OVS order (Kaiser, 2005b), OVS/Tama shows no clear

preference for either the subject or the object � but in Experiment 1, inthe same condition but now with a supportive context that renders OVS

felicitous, tama shows a significant preference for the postverbal subject.

This difference highlights the importance of investigating noncanonical

word orders in an appropriate context and also suggests that tama’s

sensitivity to word order has to do with the discourse-level/pragmatic

information conveyed by Finnish word order. By adding an appropriate

discourse context, the information structural function of OVS order is

fully realised and can thus have a clear effect on the referential propertiesof tama.

As a whole, the results of Experiment 1 support the form-specific

approach which claims that referential forms can show different degrees of

sensitivity to different factors. This asymmetrical pattern is not compatible

with either of the single-factor approaches or with a multiple-constraints

approach that assumes all forms will be equally sensitive to the same

constraints.



EXPERIMENT 2: EYE-TRACKING STUDY

In this section we report the results of a study that investigates people’s

interpretation of anaphoric expressions in a highly incremental, on-line

manner, by following their eye movements. The results of Experiment 1 show

that the final interpretation of the pronoun han is guided primarily by

syntactic role, whereas the final interpretation of the demonstrative tama is

sensitive to both linear order and syntactic role. This leaves open the

question of when, in the course of processing, these factors exert their

influence. Existing research demonstrates that eye-movements to objects in a

display are closely time-locked to potential referents that a listener is

considering as language unfolds over time (Cooper, 1974; Tanenhaus et al.,

1995; for a review see Tanenhaus & Trueswell, 2006). Thus we can use eye-

movements to shed light on what participants consider as potential referents

for pronouns and demonstratives during real-time processing.

Method

Participants. Sixteen adult native Finnish-speaking participants, mainly

students at the Helsinki University of Technology and the University of

Helsinki, took part in this experiment. They received approximately $5 for

participation in the experiment.

Procedure. An eye-movement-during-listening paradigm was employed

in which participants heard descriptions of clip-art generated pictures

(similar to Altmann & Kamide, 1999; Arnold, Eisenband, Brown-Schmidt,

& Trueswell, 2000). Participants saw large colour pictures of simple scenes

involving human or animal characters and listened to a short pre-recorded

story about each scene. Participants were told that in some cases, the story

might not match the picture, and that in such cases, their task was to correct

(by speaking out loud) the story according to what they saw in the picture.

A digital camera was used to record participants’ eye movements during

the experiment. On each trial, the participant was shown a large colour

picture, and above this picture was a SONY DVcam digital camcorder with

audio-lock recording. The DVcam camcorder was centred directly above the

picture, and recorded the participant’s face and eyes, the auditory stimuli,

and the participant’s spoken responses. The pre-recorded sound files were

played by a Dell laptop over external stereo speakers. Analysis of the eye

movements and speech onsets, described below, were done by hand on the

videotapes at a later date, using a SONY DSR-30 digital VCR with jog-

shuttle control. This video-based eye gaze process was used because the data

were collected in Finland, where neither author had access to a head-

mounted eye-tracking system.



Materials. The colour pictures used in this study were made using clip-

art and Adobe Photoshop, and were printed on 11�16 inch paper using a

high-resolution colour ink-jet printer. At a typical viewing distance of 1

metre, the visual angle of the scene subtended approximately 23 degrees.

Typically, the pictures contained two to four characters (people or animals)

as well as other objects that made up a coherent scene. Brief verbal

passages were prepared that described a simple story involving the

participants shown in the picture. The stories were spoken with neutral

intonation and recorded using the Syntrillium CoolEdit program on a

laptop PC. The same female native Finnish speaker’s voice was used for all

sound files.

There were 16 critical items (picture-story pairs) in the experiment. All

critical items contained two human characters. Both characters were

approximately the same size and positioned on opposite sides of the picture;

one on the left and one on the right. (Pilot testing was done to determine that

the characters were far enough apart to enable coders to easily distinguish

eye movements to each character from the video record.) Figure 2 presents

an example scene, which was presented with the verbal passage shown in (7).

Figure 2. Sample scene for the eye-gaze experiment.



(7)

(a) Liisa astuu sisaan eraan suuren firman paatoimistoon.

‘Liisa steps into the main office of a big company.’

(b) Han huomaa sihteerin, joka puhuu puhelimessa.

‘She notices a secretary who is talking on the phone.’

’(c) Hetken paasta sihteeri moittii juuri sisaan tullutta liikemiesta [SVO]

Moment-GEN after secretary-NOM criticises just entered businessman-PART

‘After a moment the secretary-SUBJ criticises a businessman-OBJ who has just

walked in’

(c’) Hetken paasta sihteeria moittii juuri sisaan tullut liikemies [OVS]

Moment-GEN after secretary-PART criticises just entered businessman-NOM

‘After a moment the secretary-OBJ criticises a businessman-SUBJ who has just

walked in’

(d) samalla kun printterit tulostavat paivan raportteja.

‘while the printers are churning out the day’s reports.’

(e) Han // tama seisoo valokopiokoneen lahella.

‘S/he // This is standing near a photocopier.’

Each story began with an opening sentence which describes what a

character called Liisa (not shown in the visual scenes) is doing. Then, in

the second sentence, a new referent is introduced (here, a secretary). This

referent is mentioned again in the next sentence, which has SVO or OVS

order. In both SVO and OVS conditions, the preverbal noun is discourse-

old, as it is mentioned in the preceding sentence. Thus, both SVO and OVS

sentences are felicitous. In the pictures, the position of the character to

which the preverbal noun refers was distributed between the left and right

sides of the scene.

The critical sentence (sentence (e) above) begins with either han ‘s/he’

or tama ‘this’. Crucially, this sentence is incorrect with respect to both of

the characters in the picture, since both are standing next to something

but neither is standing next to a photocopier. Participants were expected

to correct it (e.g., by saying No, s/he is standing next to a desk/a plant).

This design has two benefits: (i) it allows the experiment to be as neutral

as possible, since participants are not biased to interpret han or tama as

referring to a particular referent on the basis of preceding items, and (ii) it

allows us to collect information regarding participants’ final interpretation

of the anaphoric expression.

The sentence with the word order manipulation is separated from the

critical anaphor-initial sentence by an intervening clause which serves as a

‘look-away’ to encourage participants to look at some other object(s) in

the scene. The entities mentioned in this look-away clause are not

potential referents for han or tama due to number and/or animacy.



In addition to the 16 target items, 32 fillers were constructed. The fillers

varied in terms of the number and location of characters. In addition, the

characters used in the target items and the fillers were different, so that no

participant saw the same character twice. Twenty-four fillers were correct

and eight contained mismatches. This was done to ensure that over the entire

experiment, half of the trials were correct and half contained mismatches/

mistakes.

We constructed four presentation lists by combining the 16 target items

with the 32 fillers. All target items were separated by at least one filler.

Within a presentation list, eight target trials had OVS order and eight had

SVO order. For each order type, four of the eight trials contained the

pronoun han and four contained the demonstrative tama. Each target item

was rotated through the four conditions, which resulted in the creation of

four presentation lists. We also created reverse versions of the lists to control

for trial order.

Data analysis and coding. The videotapes of the participants’ eyes were

analysed as follows. A native Finnish speaker went through the audio

portion of each video and, using a Sony DRS-30 digital VCR (which allows

for frame-by-frame inspection of the video and audio components, at the

frequency of 30 frames per second), located the frame where the critical

sentence begins, i.e., the onset of the anaphor. (Onset coding followed the

procedure used by Snedeker, Thorpe, & Trueswell, 2001; Snedeker &

Trueswell, 2004). The video was then analysed frame-by-frame (with the

sound turned off) until the end of the trial, beginning 30 frames (1 second)

before the onset of the anaphoric expression. Coders recorded, frame-by-

frame, whether the participant was looking to the left, right, middle, or

elsewhere. Since the sound was turned off, the coders were blind to

experimental condition. The eye movement coding was used to establish

which characters participants had looked at over time, relative to the onset of

the anaphoric form. To determine the reliability of the eye gaze coding, the

first 1500 ms of the video record of three participants was double coded. The

two scorers disagreed on the left vs. right direction of gaze on less than 2% of

the video record.Our lab has used this type of eye gaze technique successfully on adult and

child participants (see Kaiser & Trueswell, 2004; Snedeker et al., 2001;

Snedeker & Trueswell, 2004). Snedeker and Trueswell (2004) discuss the

nature and validity of this kind of eye gaze technique in more depth, and

provide a detailed comparative analysis showing that a free-head video-based

eye gaze procedure produces data equivalent to that of an ISCAN head-

mounted eye-tracker. This method can be thought of as a descendent of

preferential-looking studies with children, which are known to result in very



high inter-coder reliability when frame-by-frame coding is used (Hirsh-Pasek

& Golinkoff, 1996).9

Predictions. Eye-tracking allows us to investigate the time-course of

processing and can tell us how the different sensitivities that han and tama

demonstrate in Experiment 1 unfold during the course of real-time

processing. If anaphor resolution is accomplished via a form-specific,

multiple-constraints mechanism, we predict that the asymmetric behaviour

of han and tama should be present even during the early stages of processing,

and is not merely an epiphenomenon resulting from discrete single-factor

processing. Specifically, in light of what we saw in the sentence completion

experiment, we predict that participants will show an early and persistent

preference to interpret the pronoun han as referring to the preceding subject,

regardless of whether the order is SVO or OVS. For the demonstrative tama,

we predict that when it is preceded by an SVO sentence, participants will

interpret tama as referring to the postverbal object, and that when it is

preceded by an OVS sentence, the pattern might be less clear, but

participants will nevertheless prefer the postverbal argument over the

preverbal one. In terms of linear order, our prediction is that SVO/Han is

the only condition to prompt more looks to the first-mentioned referent than

to the second-mentioned referent (i.e., exhibit a first-mention advantage).

The other three conditions � SVO/Tama, OVS/Han, and OVS/Tama � are allexpected to trigger more looks to the second-mentioned, postverbal referent

than to the first mentioned referent. Thus, we predict a word order �anaphor type interaction.

It is also worth noting that tama, unlike han, is temporally ambiguous

because in addition to its anaphoric use, it can also function as a prenominal

modifier (e.g., tama mies ‘this man’) or a discourse deictic (e.g., ‘This was

fun’). We might thus expect to see a difference in participants’ responses to

the demonstrative tama as compared to han, since it will not be clear until the

next word that tama is being used anaphorically.

For the off-line mismatch correction component of the eye-tracking

experiment, we predict that han and tama should again pattern asymme-

9 It is important to note that with this eye-gaze method, it is not possible to distinguish looks

to very close-together objects, such as the man and the plant in the sample scene. Thus, one

might be concerned that noise is introduced because looks to nearby objects may be incorrectly

coded as looks to the target characters. The presence of such noise is unlikely to distort our

results significantly, given that each character has the grammatical role of the subject in one

condition and the grammatical role of the object in another condition (see example item) and

given how objects are distributed: The 16 targets fall into four groups: object next to left

character (3 items), object next to right character (4 items), no objects next to either character (1

item) or an object next to each character (8 items). As a result, any noise resulting from looks to

other objects is predicted to affect subject entities and object entities approximately equally.



trically. However, we acknowledge that the correction task may result in

noisier data than the off-line sentence completion study in Experiment 1, due

to stronger competition from the dispreferred antecedent triggered by the

mismatch-based nature of the task (neither of the characters pictured in

the scene fits the sentence containing the anaphor). In the sentence

completion task, in contrast, there was no mismatch to trigger explicit

consideration of the other referent.

Results

The presentation of the results is divided into three main sections. First, a

general description of participants’ eye-movements is provided. Second,

detailed statistical analyses of fixation proportions to different referents over

time are reported. Third, information is provided regarding participants’ off-

line correction responses.

Overview of the eye movement patterns

Figure 3 shows the strength of the first-mentioned advantage preference

for the different conditions. The first-mentioned advantage was calculated by

subtracting the proportion of looks to the second-mentioned referent from

the proportion of looks to the first-mentioned referent, and thus a positive

number indicates more looks to the first-mentioned referent than to the

second-mentioned one, and a negative number signals that there were more

looks to the second-mentioned referent than to the first-mentioned one.

Crucially, using advantage scores allows us to take into account looks to

Figure 3. First-mentioned advantage scores for pronouns and demonstratives as a function of

time. The first-mentioned advantage score is calculated by subtracting the proportion of looks to

the second-mentioned character from the proportion of looks to the first-mentioned character.

The onset of the anaphor is at 0 ms.



either of the two referents at any point in time, which is arguably a more

complete measure than an approach that focuses only on looks to one

referent (see also Arnold et al., 2000, Arnold, Brown-Schmidt, & Trueswell,

2007, on advantage scores). However, for completeness we also provide

figures showing the proportion of looks to the first-mentioned referent and

second-mentioned referents separately (Figures 4 and 5).

SVO conditions. Figure 3 shows that for approximately the first 400 msafter the onset of the anaphor (which is at 0 ms), both SVO/Han and SVO/

Tama show a sharp rise in first-mentioned advantage scores, after which the

tama conditions show a sudden downturn away from the first-mentioned

referent. The first-mentioned advantage scores for SVO/Han, however, keep

rising until they plateau at about �0.35. These patterns can also be seen in

Figure 4 and Figure 5. For SVO/Tama, we observe a sharp increase in looks

to the second-mentioned referent (Figure 5) beginning right around the same

time as the downturn in looks to the first-mentioned referent (Figure 4). For

SVO/Han, as expected, we see that the looks to the first-mentioned referent

(Figure 4) keep rising until they reach about 0.64, whereas the looks to the

second-mentioned referent remain relatively low (Figure 5). We discuss the

SVO/Tama pattern in more detail below.

OVS conditions. Figure 3 shows that OVS/Han is almost a mirror image

of SVO/Han, and shows a strong anti-first-mention preference. Thus, both

OVS/Han and SVO/Han show a subject preference. Figures 4 and 5 show

that the rise in looks to the preceding subject begins early in both conditions.

In the OVS/Tama condition, there is clearly no first-mention (object)

Figure 4. Probability of fixating the first-mentioned referent as a function of time in each of the

four conditions.



preference (Figure 3). The first-mention scores hover around 0 until around

700 ms, at which point there is a sharp downturn towards the second-

mentioned referent, a pattern that is also evident in Figures 4 and 5. Putting

together the patterns for SVO/Tama and OVS/Tama, it becomes clear that

despite the unexpected, transient subject looks in the SVO/Tama condition,

tama is patterning very unlike han. There is no hint of a general object

preference that would complement the clear subject preference we see for

han.

Statistical analysis of eye movement patterns

To analyse the time course of the eye movement patterns in detail, we

conducted analyses of variance (ANOVAs) on seven 400 ms time-slices,

starting 800 ms before the onset of the anaphoric expression and continuing

for 2000 ms after the onset. For each time slice, participant and item means

of first-mentioned advantage scores (the proportion of looks to the first-

mentioned referent subtracted from the proportion of looks to the second-

mentioned referent) were entered into separate ANOVAs with four factors:

Word order (SVO or OVS), Anaphor type (han or tama), Order (forward or

reverse list), and List (4 levels) in the participant analysis and item group (4

groups) in the item analysis.10

Figure 5. Probability of fixating the second-mentioned referent as a function of time in each of

the four conditions.

10 Whenever an ANOVA was conducted on an advantage score calculated on the basis of

proportions of looks, a parallel ANOVAwas conducted on an arcsine transformation of the data

(arcsin (sqrt(x))). This was done to adjust for the fact that a proportion is bounded at 0 and 1.

Throughout the paper, we report the F values, p values and means from the untransformed data.

Unless otherwise noted, any effect that was reliable in the untransformed data was also reliable

in the transformed data.



During the first two time slices (�800ms to �400ms, �400ms to 0 ms,

anaphor onset), there are no significant effects of anaphor type or word

order in the subjects or the items analyses. However, there is a marginal

effect of word order during the second time slice in the items analysis,

F1(1, 8)�1.73, p�.225, F2(1, 12)�3.8, p�.075, with SVO exhibiting

a stronger first-mention advantage score than OVS. This effect is

also marginal in the items analysis of the arcsine transformed data,

F2(1, 12)�3.589, p�.082.

During the third time slice (0�400 ms, where 0 ms refers to anaphor onset),

we see an effect of word order that is significant by items and marginal by

subjects F1(1, 8)�3.8, p�.087; F2(1, 12)�9.51, pB.01. [In the arcsine

transformed data, the word order effect is significant by both subjects and

items, F1(1, 8)�5.632, pB.05, F2(1, 12)�9.086, pB.05.] As Figure 3 shows,

during this time slice, first-mentioned advantage scores are higher with SVO

than OVS order, as people are looking more at the subject with SVO than

with OVS. There are no effects of anaphor type and no interaction.

During the fourth time slice (400�800 ms), the word order effect persists,

F1(1, 8)�7.02, pB.05; F2(1, 12)�38.33, pB.01, and there is also a

significant word order-anaphor type interaction, F1(1, 8)�7.21, pB.05;

F2(1, 12)�6.16, pB.05. There are no significant main effects of anaphor

type. As can be seen in Figure 3, it is during this time slice that SVO/Tama

separates from SVO/Han, and takes a sudden plunge towards the second-

mentioned referent. The difference in the first-mentioned advantage scores in

the SVO/Han and the SVO/Tama conditions thus becomes bigger than the

corresponding difference in the OVS/Han and OVS/Tama conditions, which

shows up as a significant interaction.

This pattern persists in the fifth time slice (800�1200 ms), where there is a

significant main effect of word order, F1(1, 8)�11.09, p�.01; F2(1, 12)�26.95, pB.01, with SVO resulting in a higher first-mention advantage score

than OVS. There is also a significant word order-anaphor type interaction

due to the fact that only SVO/Han triggers a high proportion of looks to the

first-mentioned referent, F1(1, 8)�8.62, pB.05; F2(1, 12)�16.33, pB.01.

There are no significant effects of anaphor type.

Similarly, in the sixth time slice (1200�1600 ms, we see a continuing main

effect of word order, F1(1, 8)�13.67, pB.01; F2(1, 12)�15.7, pB.01, and

an anaphor-word order interaction that is marginal by subjects and

significant by items, F1(1, 8)�3.8, p�.087; F2(1, 12)�4.76, p�.05. There

is also a marginal main effect of anaphor type, F1(1, 8)�4.786,

p�.061, F2(1, 12)�3.238, p�.097, but the presence of the word order-

anaphor interaction indicates that it is primarily the SVO/Han condition

that is triggering a high proportion of looks to the first-mentioned referent.



The word order-anaphor type interaction is also significant in the final time

slice (1600�2000 ms) by subjects, F1(1, 8)�5.56, pB.05, but not by items

(p�.146).11

Planned comparisons were conducted on the first-mentioned advantage

scores (looks to first mentioned minus looks to second mentioned

referent) and subject-advantage scores (looks to subject minus looks to

object) of han and tama for each time slice. The analyses reveal that in the

han conditions, the first-mentioned advantage scores are influenced

significantly by word order in the third, fourth, fifth and sixth time

slices: third time slice: F1(1, 8)�5.67, pB.05; F2(1, 12)�6.86, pB.05;

fourth time slice: F1(1, 8)�11.37, p�.01; F2(1, 12)�33.78, pB.01; fifth

time slice: F1(1, 8)�6.38, pB.05; F2(1, 12)�22.45, pB.01; sixth time

slice: F1(1, 8)�6.4, pB.05; F2(1, 12)�14.19, pB.01.12 In other words, in

the han conditions, OVS order triggers a significantly lower proportion of

looks to the first-mentioned referent than SVO order, starting less than

400 ms after the onset of the anaphor. In contrast, word order has no

significant effect during any time slice on the subject advantage scores for

the pronoun.

In the tama conditions, the first-mentioned advantage scores are not

significantly influenced by word order at any point in time, with the

exception of a marginal effect of word order during the third time slice in

the items analysis of the arcsine transformed data F2(1, 12)�3.589, p�0.083. It is during this time slice that looks to the subject (first-mentioned

referent) are still increasing in the SVO/Tama condition, before the sharp

downturn towards the (second-mentioned) object. After this point, SVO

and OVS both induce comparable proportions of looks to the first vs.

second-mentioned referents. Turning to the subject advantage scores, we

see a significant effect of word order in the sixth time slice (1200�1600 ms,

F1(1, 8)�7.06, pB.05; F2(1, 12)�5.29, pB.05. In that segment, there are

significantly more looks to the (second-mentioned) subject in OVS order

than to the (first-mentioned) subject in SVO order. The subject advantage

scores for other time slices do not show significant effects of word order,

indicating that the (un)likelihood of tama to refer to the subject is not

significantly influenced by word order, except during the 1200�1600 ms

time slice.

11 Analyses of the arcsine transformed data show a trend towards a word order-anaphor type

interaction by subjects in the final time slice, F1(1, 8)�3.55, p�.096, but not by items (p�.118).12 In the arcsine transformed data, there is also a marginal effect of word order in the final

time slice in the subjects analysis, but not in the items analysis, F1(1, 8)�5.105, p�.054, F2(1,

12)�2.286, p�.156.



Off-line referential judgements

As shown in Figure 6, in the SVO/Han condition, 81% of the responses

treated the pronoun as referring to the preceding subject.13 In the OVS/Han

condition, 62% of the responses treated the preceding subject as the

antecedent of the pronoun. Although participants are more likely to

interpret han as referring to the subject than the object in both SVO and

OVS conditions, the strength of the subject preference is not equally strong

with SVO and OVS, in contrast to Experiment 1.

The results for the demonstrative conditions show that, as we already saw

in the eye-movement data and in Experiment 1, tama does not exhibit an

overwhelming sensitivity to the syntactic role of the antecedent. With OVS/

Tama, participants show a clear bias to interpret the second-mentioned,

postverbal referent as the antecedent of the demonstrative (76% corrections

towards the postverbal subject). In the SVO/Tama condition, 54% of

responses treated the second-mentioned referent as the antecedent of the

demonstrative. Although postverbal interpretations are more frequent in

both SVO/Tama and OVS/Tama conditions, the rate of postverbal object

choices in SVO/Tama is lower than expected on the basis of Experiment 1.

For data analysis, first-mentioned advantage scores were calculated by

subtracting the proportion of corrections towards the second-mentioned

referent from the proportion of corrections towards the first-mentioned

referent. Participant and item means of first-mentioned advantage scores

were entered into separate ANOVAs with four factors: Word order (SVO or

OVS), Anaphor type (han or tama), Order (forward or reverse list), and List

Figure 6. Percentage of corrections towards the first-mentioned referent and the second-

mentioned referent for pronouns and demonstratives. (Ambiguous responses are excluded.)

13 Some of the corrections were ambiguous (e.g., ‘No one is standing next to a photocopier’).

The percentages reported here exclude these ambiguous responses. However, the patterns do not

change even if these responses are included, because the proportion of ambiguous responses in

the four conditions is very similar.



(four levels) in the participant analysis and item group (four groups) in the

item analysis.

The ANOVAs show that the first-mentioned advantage scores are

significantly influenced by anaphor type F1(1, 8)�10.44, p�.012, F2(1,

12)�12.83, pB.01, and by word order F1(1, 8)�12.76, pB.01, F2(1, 12)�53.7, pB.001). Thus, overall the han conditions result in more corrections

towards the first-mentioned referent (as compared to the second-mentioned

referent) than the tama conditions, and SVO order triggers more corrections

towards the first-mentioned referent than OVS order. There is also a

marginal anaphor type � word order interaction F1(1, 8)�3.96, p�.08,

F2(1, 12)�3.85, p�.07, which is slightly stronger when the analyses are

conducted on arcsine-transformed data, F1(1, 8)�4.77, p�.06, F2(1, 12)�3.85, p�.07. As Figure 6 shows, the strength of the preference for the first-

mentioned referent relative to the second-mentioned referent is more

influenced by word order (SVO vs. OVS) in the han conditions than in the

tama conditions.

Discussion

As a whole, the results of the eye-tracking study shows that participants’

interpretation of han and tama is influenced by the antecedent’s position

(pre- vs. postverbal) and syntactic role (subject vs. object) to different

degrees. The strong subject preference we observe in participants’ eye-

movements in the han conditions, combined with the absence of a

corresponding overarching object preference for tama, is compatible with

the form-specific multiple-constraints approach, but not with either of the

single-factor approaches or by a multiple-constraints approach that assumes

all forms are equally sensitive to the same constraints.

As the eye-tracking results show, the asymmetrical behaviour of han and

tama emerges early during processing. The pronoun han shows a sensitivity

to syntactic role beginning shortly after the onset of the anaphor, with SVO/

Han showing a steep rise in first-mention advantage scores and OVS/Han

showing an early increase in looks to the second-mentioned referent. Even

though the eye-movement data for the demonstrative tama are less clear, they

crucially reveal a striking asymmetry relative to han, since there is no

overarching syntactic role sensitivity with tama.

The eye-movements in the SVO/Tama condition show a temporary initial

rise in looks to the subject, which we suggest may be a garden-path effect

made possible by two properties of Finnish coinciding: (i) SVO order appears

to create the expectation that the subject will be mentioned again in the next

sentence, and (ii) tama can function as a prenominal modifier (e.g., ‘this

man’). As can be seen in Figure 3, SVO conditions have a slightly higher

first-mentioned advantage than OVS conditions even before the anaphor. As



it turns out, this is what Centering Theory (Grosz, Joshi, & Weinstein, 1995,

inter alia) predicts: A situation in which a preceding discourse-old subject is

mentioned again in subject position is claimed to be easy to process and thus

predicted to occur frequently (see also Givon, 1983; Arnold, 1998; Prince,

1992). In other words, a discourse-old referent in subject position is likely to

be mentioned again in subject position. The pronoun han allows the system

to fulfil this expectation, but the demonstrative tama does so only if tama is

parsed as a modifier referring to the preceding subject. Indeed, participants’

eye-movements suggest that they are initially interpreting tama as modifying

a noun referring to the preceding subject, e.g., ‘this man’.

The idea that the transient early subject looks in the SVO/Tama condition

are due to a word-order driven expectation combined with tama interpreted

temporarily as a modifier is compatible with the duration of the anaphor.

The average duration of the demonstrative was about 230 ms and the average

duration of the pronoun was 180 ms. The sound files were measured using

Praat software (Paul Boersma and David Weenink, Institute of Phonetic

Sciences, University of Amsterdam). The duration of tama � combined with

the fact that it takes about 150�200 ms to launch and program an eye

movement (see e.g., Matin, Shao, & Boff 1993; see also Altmann & Kamide,

2004; Dahan, Magnuson, Tanenhaus, & Hogan, 2001; Saslow, 1967) � is

compatible with the timing of the downturn, which starts around 400 ms

after the onset of the anaphor. This suggests that once it becomes clear that

tama is not modifying a subsequent noun but rather acting as a free-standing

anaphoric form, the parser realises its mistake and this triggers looks away

from the subject.Even after it has become clear that tama is being used anaphorically, the

strength of the object preference demonstrated by tama in the SVO/Tama

condition is not as strong as we might expect on the basis of Experiment 1.

We hypothesise that this stems at least partially from a lingering effect of the

garden path. As Christianson et al. (2001) observed, even after a sentence

has been re-analysed, aspects of the initial parse can still persist, i.e., the

initial interpretation is not necessarily fully eradicated (see also Tabor,

Galantucci, & Richardson, 2004). In our case, this would mean that the

interpretation of tama as modifying a noun referring to the subject would

linger, and weaken the predicted object preference of anaphoric tama.

The main effect of word order that we found in participants’ eye-

movements, indicative of a bigger first-mentioned advantage with SVO order

than OVS order, is another sign that SVO/Tama is not patterning as we

expected. The prediction was that only SVO/Han would show a clear first-

mentioned advantage and all other conditions would show a clear second-

mentioned advantage, resulting in a word order � anaphor interaction and

no main effect of word order.



Nevertheless, if we consider the eye-movement patterns in all four

conditions as a whole, the asymmetry between han and tama is clear. Even

though SVO/Tama does not pattern quite as we predicted, it clearly fails to

fit the predictions of either the single-factor approach or a multiple-constraints approach that assumes all forms share the same sensitivities.

Thus, the results are compatible with our hypothesis that han and tama differ

in how sensitive they are to different kinds of information.

Participants’ off-line corrections are also compatible with the claim that

han and tama differ in how sensitive they are to the linear position and

syntactic role of potential antecedents. Other approaches are unable to

account for the data patterns satisfactorily: The corrections in the SVO/Han

condition show a subject preference and those in the OVS/Tama conditionalso have a preference for the (now postverbal) subject � a pattern which is

not compatible with an account based only on the effects of syntactic role.

Moreover, when this pattern is combined with the absence of an object

preference in OVS/Han condition, it also becomes clear that a pure linear

order account is not supported by the data. Furthermore, a multiple-

constraints account according to which syntactic role and linear order are

weighted equally cannot explain the subject preference in the OVS/Tama

condition.However, the patterns in the off-line correction data are not as clear as

in the sentence completion patterns of Experiment 1. This may be due to

(i) the hypothesised garden-pathing in the SVO/Tama condition and (ii)

stronger competition from the dispreferred antecedent, as discussed in the

Predictions section of Experiment 2. More specifically, it is possible that

the difference in results between the completion data in Experiment 1 and

the correction data in Experiment 2 reflects something about the tasks. In

Experiment 2, the sentence containing the critical anaphor does not matcheither of the pictured characters (e.g., neither is standing next to a

photocopier), and the participant � in correcting the sentence-scene

mismatch � must report some accommodation of the sentence that would

make it match the scene. This might introduce participant-specific

strategies that are unrelated to pronoun resolution, which would in turn

make the results noisier than those seen in the sentence completion

experiment. Note for instance that the order and direction of all effects are

the same across both experiments (compare Figures 1 and 6) but thepattern is simply less clear in Experiment 2. Nevertheless, it is also worth

noting that in the off-line correction data, the subject preference is not

equally strong with SVO/Han and OVS/Han, a finding that seems

compatible with claims made by Jarvikivi et al. (2005).

The fact that tama is temporally ambiguous whereas han is not raises the

question of whether the asymmetrical behaviour of the demonstrative and

pronominal forms could stem from this difference. Do the different



referential preferences of han and tama stem from the fact that one form is

temporally ambiguous and the other is not?

In our opinion, the ambiguous/unambiguous distinction could not

generate the asymmetrical sensitivity to word order and grammatical role

that we observe for han and tama. It seems like the ambiguity of tama,

combined with the expectations triggered by SVO order, results in the SVO/

Tama condition initially patterning just like SVO/Han, i.e., showing a

subject/first-mentioned preference. As a result, SVO/Tama and SVO/Han

initially group together by preferring the first-mentioned referent, and OVS/

Han and OVS/Tama resemble each other in preferring the second-mentioned

referent over the first-mentioned one. Thus, if anything, the garden-pathing

(which makes SVO/Tama initially pattern like SVO/Han) should prevent the

word order � anaphor type interaction from emerging. Our finding that a

significant word order � anaphor type interaction (with tama being more

sensitive to word order than han) emerges after the hypothesised garden-

pathing does not seem to be derivable from � in fact, seems to go against the

consequences of � the ambiguity of tama.14 Our results would be stronger if

the garden-pathing were not present.Furthermore, even though we do not believe this to be the case for our

results, it would be theoretically and empirically interesting if temporally

ambiguous anaphoric forms show a fundamentally different behaviour from

unambiguous forms. A number of languages resemble Finnish in allowing

demonstratives to be used both anaphorically for human referents and in

prenominal modifier position (e.g., German and Dutch). Even English ‘that’

contrasts with ‘it’, since ‘that’ can occur on its own (‘Could you put that on

the table?’) as well as prenominally (‘that book’). Finding that temporary

ambiguity results in these forms patterning differently from unambiguous

pronouns would offer an interesting explanation for why the forms pattern

differently, and would not necessarily be incompatible with the form-specific

account.

In sum, Experiment 2 corroborates the lack of a single unifying factor

behind han and tama; they pattern distinctly, as predicted by the form-

specific approach. The eye-tracking results show that even on an incremental

level, we cannot maintain an approach which assumes that pronouns and

demonstratives are sensitive to a single notion of salience. Han and tama

should not display different sensitivities to different factors if what they

correspond to are simply two different rankings on a one-dimensional

salience hierarchy.

14 One could, of course, try to connect ambiguity and referential properties by stipulating

that temporarily ambiguous referential forms are more sensitive to word order than

unambiguous forms. However, we do not pursue this possibility because we have not

encountered any evidence that would support or motivate this claim.



GENERAL DISCUSSION

The main aim of this research was to study the referential properties of

pronouns and demonstratives in Finnish, in order to investigate the validity

of the assumption that all referential forms can be ranked along a unified

salience scale. The work more generally was aimed at improving our

understanding of what it means for a referent to be salient. We wanted to

test the validity of the form-specific multiple-constraints approach to

reference resolution, which was supported by earlier off-line results for

Finnish pronouns and demonstratives (Kaiser, 2003, 2005b). The sentence

completion experiment and the eye-tracking study presented in this paper

indicate that the pronoun han ‘s/he’ and the demonstrative tama ‘this’ are

sensitive to different factors, as predicted by the form-specific approach. In

the kinds of contexts investigated here, han is interpreted as referring to

preceding subjects, whereas tama prefers postverbal, discourse-new referents,

especially objects. Thus, whereas the interpretation of the pronoun han is

driven primarily by the syntactic role of potential antecedents, the

demonstrative tama exhibits a sensitivity to both word order/information

structure and syntactic role. These results are problematic for an approach

that treats han and tama as being sensitive to the same ‘kind’ of salience

(whether it be determined by a single factor, such as syntactic role or

information structure, or by a set of factors). Rather, they provide support

for a multi-dimensional approach, where anaphoric forms can be sensitive to

different factors to different degrees. In this form-specific approach, each

anaphoric form has its own set of weighted constraints that guide its

interpretation.

It is worth contrasting this approach with a more extreme interpretation:

namely that han is sensitive only to syntactic information and only tama is

sensitive to more than one kind of information. We do not want to make

such a strong claim, because we think it likely that all anaphoric forms are

sensitive to more than one constraint, but to varying degrees. In the case of

han, our view is that Finnish listeners are implicitly aware that the

grammatical roles of preceding entities are of primary relevance for

computing the referent of han, but this does not preclude using other

information sources. For example, han can be used without a linguistic

antecedent if the referent is sufficiently salient in the extra-linguistic context.

Consider a situation where an unknown person just rode by very quickly on

a bicycle. In this situation, a speaker could exclaim ‘Boy, was he fast!’, and

use han without a linguistic antecedent, without making reference to a

preceding subject. A second example of another information source having

an effect on the interpretation of han comes from repeated occurrences of a

pronoun. Kaiser (2003, 2005a) found that the referential form of the

preceding subject and object (full NP vs. pronoun) can have an effect on



the interpretation of a subsequent pronoun. She hypothesises that a chain of

pronouns patterns differently from the initial use of han to refer to a full NP

(see also Kameyama, 1999; Beaver, 2004), and proposes a referent-tracking

system that builds on this distinction. Thus, we suggest that even if a form isprimarily sensitive to a particular constraint, this does not prevent other

constraints from also playing some role in the reference computation process.

The form-specific multiple-constraints approach also receives support

from research on Estonian pronouns and demonstratives (Kaiser & Vihman

2006; 2008). Estonian, a Finno-Ugric language with flexible word order and

closely related to Finnish, can also use pronouns (ta ‘s/he’) and demonstra-

tives (see ‘this’) to refer back to third-person human antecedents. A sentence

completion study, modelled on Experiment 1, shows that the Estonianpronoun ta ‘s/he’ is sensitive to the syntactic role of its antecedent, whereas

see ‘this’ is sensitive to both syntactic role and word order. In both Finnish

and Estonian, pronominal anaphors show a primary sensitivity to the

syntactic role of the antecedent, but the demonstrative anaphors are not

‘mirror images’ of the pronouns � i.e., they do not simply refer to entities

with lower-ranked syntactic roles.

To investigate the form-specific multiple-constraints approach in the

within-sentence domain, Kaiser, Runner, Sussman, and Tanenhaus (inpress-a, in press-b) investigated the interpretation of English pronouns and

reflexives. The results suggest that pronouns and demonstratives in sentence-

internal contexts such as Peter told Andrew about the picture of himself/him

differ in the degree of sensitivity they exhibit to structural and non-structural

factors, further supporting the form-specific approach.

The form-specific multiple-constraints approach is compatible with the

growing body of research indicating that reference resolution is influenced by

multiple factors (see e.g., Ariel, 1990; Arnold, 1998; Arnold et al., 2000;Badecker & Straub, 2002). Subsequent to the completion of the experiments

reported in this paper, Jarvikivi, van Gompel, Hyona, and Bertram (2005)

also reported results from a visual-world eye-tracking study investigating the

interpretation of the pronoun han when preceded by SVO and OVS order. In

a visual-world eye-tracking study with SVO and OVS sentences, Jarvikivi et

al. found that han exhibits an initial sensitivity to syntactic role (preferring

subjects) 480 ms to 690 ms after the onset of the pronoun, followed by effects

of both syntactic role and word order from 690 ms onwards. Theirexperimental design differs somewhat from ours, especially in that they did

not have context sentences preceding the critical SVO/OVS sentences. As

discussed in conjunction with the review of the Kaiser (2005b) results,

previous work on Finnish has shown that OVS order, if presented out of

context, causes significant processing difficulties, unlike SVO order which is

felicitous in an all-new context. It is not clear how this felicity asymmetry

between SVO and OVS may have affected Jarvikivi et al.’s results. As a whole,



although the results of Jarvikivi et al. (2005) differ somewhat from ours, in

our opinion they are not incompatible with our main claim, namely that the

pronoun han and the demonstrative tama differ in how sensitive they are to

different types of information.

Given our claim that our findings cannot be reconciled with a traditional

view of salience, should the notion of salience be discarded? In our opinion,

one should not abandon the basic observation that more informative

referential forms (e.g., ‘the man with the straw hat’, ‘that man’) can be

used to refer to less salient referents than informationally impoverished

forms (e.g., ‘he’) (Givon, 1983; Ariel, 1990; Gundel et al., 1993, inter alia).

The implications of our findings are most relevant for languages that have

two (or more) anaphoric forms that cannot be distinguished on the basis of

their informativity, such as Finnish han and tama or English it and this/that

(see Ariel 2001, p. 29). Moreover, even when these forms display asymmetric

sensitivities, one could potentially pursue an alternative approach that

incorporates the notion of salience somewhat more indirectly, namely the

idea that han and tama cause comprehenders to preferentially probe different

types of representations when trying to locate the most likely referent for

each form. The different sensitivities that han and tama exhibit to word order

and syntactic role are, according to this view, reflexes of a more fundamental

difference between the two referential forms, namely the representational

level on which they are resolved. According to this approach, when a

comprehender processes the sentence containing han or tama, two repre-

sentations of the prior linguistic input remain activated and are relevant for

anaphor resolution:

a. the syntactico-semantic representation of the preceding sentence, which

we assume includes information about grammatical and thematic roles.

b. the comprehender’s mental model of the discourse, which we assumeincludes information about the situation or event being described and

the entities involved in it. In our view, this mental discourse model is

not a representation of a text (see Glenberg, Kruley, & Langston, 1994)

but rather a discourse model that the comprehender constructed on the

basis of the preceding discourse (see Johnson-Laird, 1983; Van Dijk &

Kintsch, 1983).

Existing work supports the idea that these two types of representation are

accessed during processing. For example, research on verb-phrase ellipsis

suggests that a syntactico-semantic representation of the preceding sentences

needs to be maintained, at least temporarily (see Fiengo & May, 1994;

Shapiro & Hestvik, 1995; Shapiro, Hestvik, Lesan, & Garcia, 2003). The role

of various types of mental models has been investigated by a range of



researchers, including Van Dijk and Kintsch (1983), Glenberg et al. (1994)

and Johnson-Laird (1983).

This approach would allow us to maintain the notion of salience by

hypothesising that the relevant entities are ranked in terms of their salienceon the syntactico-semantic level as well as on the mental discourse model

level. In light of existing research, it seems reasonable to hypothesise that (a)

on the syntactico-semantic level, agentive subjects are more salient than non-

agentive objects and oblique arguments, and (b) on the level of the mental

representation of the discourse, salience is influenced by a range of factors, in

particular information-structural factors. Thus, under this approach han

would be described as preferring referents that are salient on the syntactico-

semantic level and tama as preferring referents that are lower in salience onthe discourse model level.

The asymmetrical sensitivities of han and tama raise the question of why

referential forms differ in the kind of information they are most sensitive to.

Why is tama ‘this’ more sensitive to word order and information structure

than han ‘s/he’? We do not offer a definitive answer, but note that the

difference may be related to the generally discourse-bound nature of tama. In

addition to functioning as an anaphor for human referents, this form is also

used as a proximal demonstrative and a discourse deictic. In these uses, thereferent of tama is extremely context-dependent and often does not have an

antecedent that is a linguistic constituent. This contrasts with the pronoun

han, which is used to refer to concrete human entities. In light of these

differences, the finding that syntactic role does not play as important a role

for tama as it does for han no longer seems very surprising.

The finding that syntactic role is not the primary determinant of a

demonstrative’s antecedent is also supported by eye-tracking work by

Brown-Schmidt et al. (2004, 2005) investigating it and that in English.They found that both it and that are sensitive to extra-linguistic information,

such as how easily two objects could be viewed as a composite. For instance,

given a command like ‘Put the cup on the saucer. Now put that over by the

shovel’, participants interpreted that as referring to the composite ‘cup-and-

saucer’ 88% of the time. Perhaps the preference of that to refer to composite

entities is related to the information-structural sensitivity of tama, another

form that can be used to refer to entities that do not have linguistic

antecedents.In sum, our investigation of the referential properties of the pronoun han

‘s/he’ and the demonstrative tama ‘this’ in Finnish suggests that not all

referential forms within a single language are sensitive to the same salience-

influencing factors to the same degree. As our results show, han and tama

differ in the degree of sensitivity they exhibit to the syntactic role and linear

position of potential antecedents � a finding which is not compatible with

single-factor approaches to reference resolution, nor with a multiple-factor



approach assuming that all referential forms are equally sensitive to different

kinds of information. We interpret our results as support for the form-

specific multiple-constraints approach, and we also explore the possibility

that the different referential properties of han and tama are due to these

forms prompting comprehenders to preferentially probe different sorts of

representations when trying to locate the most likely referent within the

discourse.

Manuscript received September 2006

Revised manuscript received October 2007

First published online May 2008

REFERENCES

Albrecht, J. E., & Clifton, C., Jr. (1998). Accessing singular antecedents in conjoined phrases.

Memory and Cognition, 26, 599�610.Altmann, G. T. M., & Kamide, Y. (2004). Now you see it, now you don’t: Mediating the mapping

between language and the visual world. In J. M. Henderson & F. Ferreira (Eds.), The interface of

language, vision, and action: Eye movements and the visual world (pp. 347�386). New York:

Psychology Press.

Ariel, M. (1990). Accessing NP antecedents. London: Routledge, Croom Helm.

Ariel, M. (2001). Accessibility theory: an overview. In T. Sanders, J. Schilperoord, & W. Spooren

(Eds.), Text representation, linguistic and psycholinguistic aspects (pp. 29�87). Amsterdam/

Philadelphia: John Benjamins.

Arnold, J. (1998). Reference form and discourse patterns. Ph.D. dissertation, Stanford University,

Stanford, CA.

Arnold, J. E., Brown-Schmidt, S., & Trueswell, J. C. (2007). Children’s use of gender and order-of-

mention during pronoun comprehension. Language and Cognitive Processes, 22, 527�565.Arnold, J. E., Eisenband, J. G., Brown-Schmidt, S., & Trueswell, J. C. (2000). The rapid use of

gender information: Evidence of the time course of pronoun resolution from eyetracking.

Cognition, 76, B13�B26.Badecker, W., & Straub, K. (2002). The processing role of structural constraints on the

interpretation of pronouns and anaphors. Journal of Experimental Psychology: Learning,

Memory and Cognition, 28, 748�769.Beaver, D. (2004). The optimization of discourse anaphora. Linguistics and Philosophy, 27(1), 3�56.Brennan, S., Friedman, M., & Pollard, C. (1987). A centering approach to pronouns. Proceedings of

the 25th Annual Meeting of the Association for Computational Linguistics (pp. 155�162).Stanford, CA.

SOURCES

Internet, websites of the newspapers Aamulehti and Viitasaaren Seutu.

Linna, V. (1954/1999). Tuntematon Sotilas. Helsinki: Werner Soderstrom Oy.

Remes, I. (2000). Paakallokehraaja. Helsinki: Werner Soderstrom Oy.



Brown-Schmidt, S., Byron, D. K, & Tanenhaus, M. (2005). Beyond salience: Interpretation of

personal and demonstrative pronouns. Journal of Memory and Language, 53, 292�313.Brown-Schmidt, S. B., Byron, D., & Tanenhaus, M. K. (2004). That’s not it and ‘it’s’ not ‘that’: The

role of conceptual composites in in-line reference resolution. In M. Carreiras & C. Clifton, Jr.

(Eds.), On-line sentence processing: ERPS, eye movements and beyond. Hove, UK: Psychology

Press.

Chafe, W. L. (1976). Givenness, contrastiveness, definiteness, subjects, topics, and point of view. In

C. N. Li (Ed.), Subject and topic (pp. 25�55). New York: Academic Press.

Chesterman, A. (1991). On definiteness. Cambridge, UK: Cambridge University Press.

Christianson, K., Hollingworth, A., Halliwell, J. F., & Ferreira, F. (2001). Thematic roles assigned

along the garden path linger. Cognitive psychology, 42, 368�407.Cooper, R. M. (1974). The control of eye fixation by the meaning of spoken language: A new

methodology for the real-time investigation of speech perception, memory and language

processing. Cognitive Psychology, 6, 84�107.Crawley, R., & Stevenson, R. (1990). Reference in single sentences and in texts. Journal of

Psycholinguistic Research, 19(3), 191�210.Dahan, D., Magnuson, J. S., Tanenhaus, M. K., & Hogan, E. M. (2001). Subcategorical

mismatches and the time course of lexical access: Evidence for lexical competition. Language

and Cognitive Processes, 16, 507�534.Di Eugenio, B. (1998). Centering in Italian. In M. A. Walker, A. K. Joshi, & E. F. Prince (Eds.),

Centering theory in discourse (pp. 115�138). Oxford: Oxford University Press.

Etelamaki, M. (1996). Keskutelu tarkoitteesta kamerataiteen oppitunnilla � ja pronominit tuo, se,

tama. Thesis, Department of Finnish, University of Helsinki.

Fiengo, R., & May, R. (1994). Indices and identity. Cambridge, MA: MIT Press.

Garnham, A. (2001). Mental models and the interpretation of anaphora. Hove, UK: Psychology

Press.

Garrod, S. C., & Sanford, A. J. (1982). The mental representation of discourse in a focused memory

system: Implications for the interpretation of anaphoric noun phrases. Journal of Semantics, 1,

21�41.Gernsbacher, M. A., & Hargreaves, D. (1988). Accessing sentence participants: The advantage of

first mention. Journal of Memory and Language, 27, 699�717.Givon, T. (1983). Topic continuity in discourse: A quantitative cross-language study. Philadelphia/

Amsterdam: John Benjamins.

Glenberg, A., Kruley, P., & Langston, W. E. (1994). Analogical processes in comprehension. In M.

A. Gernsbacher (Ed.), Handbook of psycholinguistics (pp. 609�640). San Diego, CA: Academic

Press.

Gordon, P. C., & Chan, D. (1995). Pronouns, passives and discourse coherence. Journal of Memory

and Language, 34, 216�231.Gordon, P. C., Grosz, B., & Gilliom, L. (1993). Pronouns, names, and the centering of attention in

discourse. Cognitive Science, 17, 311�347.Greene, S., McKoon, G., & Ratcliff, R. (1992). The role of implicit causality and gender cue in the

interpretation of pronouns. Language and Cognitive Processes, 73(4), 231�255.Grosz, B., Joshi, A., & Weinstein, S. (1995). Centering: A framework for modeling the local

coherence of discourse. Computational Linguistics, 21(2), 203�225.Gundel, J., Hedberg, N., & Zacharski, R. (1993). Cognitive status and the form of referring

expressions in discourse. Language, 69, 274�307.Hajicova, E., & Vrbova, J. (1982). On the role of the hierarchy of activation in the process of natural

language understanding. In J. Horecky (Ed.), COLING 82 (pp. 107�113). Amsterdam: North

Holland.



Hajicova, E., Kubon, V., & Kubon, P. (1990). Hierarchy of salience and discourse analysis and

production. Proceedings of the 13th Conference on Computational linguistics, 3, 144�148.Helsinki, Finland.

Halmari, H. (1994). On accessibility and coreference. Nordic Journal of Linguistics, 17, 35�59.Helasvuo, M.-L. (2001). Syntax in the making: The emergence of syntactic units in Finnish

conversation. Philadelphia/ Amsterdam: John Benjamins.

Hiirikoski, J. (1995). Correlations between some morphosyntactic features and word order in

Finnish and English: Some preliminary results of testing the transitivity hypothesis. In B.

Warwik, S. K. Tanskanen, & R. Hiltunen (Eds.), Organization in discourse, proceedings from the

Turku Conference, Anglicana Turkuensia, 14, 289�299.Hirsh-Pasek, K., & Golinkoff, R. M. (1996). The intermodal preferential looking paradigm: A

window onto emerging language comprehension. In D. McDaniel & C. McKee (Eds.), Methods

for assessing children’s syntax. Language, speech, and communication (pp. 105�124). Cambridge,

MA: MIT Press.

Hoffman, B. (1998). Word order, information structure and centering in Turkish. In M. A. Walker,

A. K. Joshi & E. F. Prince (Eds.), Centering theory in discourse (pp. 251�272). Oxford: Oxford

University Press.

Hyona, J., & Hujanen, H. (1997). Effects of case marking and word order on sentence parsing in

Finnish: an eye fixation analysis. Quarterly Journal of Experimental Psychology, 50A(4), 841�858.

Iivonen, A. (1998). Intonation in Finnish. In D. Hirst & A. D. Cristo (Eds.), Intonation systems: A

survey of twenty languages. Cambridge, UK: Cambridge University Press.

Jarvikivi, J., van Gompel, R. P. G., Hyona, J., & Bertram, R. (2005). Ambiguous pronoun

resolution: Contrasting the first-mention and subject-preference accounts. Psychological

Science, 4, 260�264.Johnson-Laird, P. (1983). Mental models: Towards a cognitive science of language, inference and

consciousness. Cambridge, MA: Harvard University Press.

Kaiser, E. (2000). Pronouns and demonstratives in Finnish: Indicators of referent salience. In P.

Baker, A. Hardie, T. McEnery, & A. Siewierska (Eds.), Proceedings of the Discourse Anaphora

and Reference Resolution Conference (DAARC 2000), University Center for Computer Corpus

Research on Language, Technical Papers Vol. 12 (pp. 20�27). Lancaster, UK: University of

Lancaster.

Kaiser, E. (2003). The quest for a referent: A crosslinguistic look at reference resolution. Ph.D.

dissertation, University of Pennsylvania, Philadelphia, PA.

Kaiser, E. (2005a). Different forms have different referential properties: Implications for the notion

of ‘salience’. In A. Branco, T. McEnery, & R. Mitkov (Eds.), Anaphora processing: Linguistic,

cognitive and computational modelling (pp. 261�282). Philadelphia/Amsterdam: John Benjamins.

Kaiser, E. (2005b). When salience isn’t enough: Pronouns, demonstratives and the quest for an

antecedent. In R. Laury (Ed.), Minimal reference in Finnic: The use and interpretation of

pronouns and zero in Finnish and Estonian discourse (pp. 135�162). Helsinki, Finland:

Suomalaisen Kirjallisuuden Seura.

Kaiser, E. (in press). (Anti)logophoricity in Finnish. In The Proceedings from the Main Session of

the Chicago Linguistic Society’s Fortieth Meeting. Volume 40-1. Chicago, IL: Chicago Linguistic

Society.

Kaiser, E., Runner, J., Sussman, R., & Tanenhaus, M. (in press-a). The interpretation of pronouns

and reflexives in picture noun phrases: Effects of non-structural factors. In A. Alcazar, R.

Mayoral Hernandez, & M. T. Martinez (Eds.), Proceedings of the Western Conference on

Linguistics 2004. Fresno, CA: California State University at Fresno.

Kaiser, E., Runner, J., Sussman, R., & Tanenhaus, M. (in press-b). The real-time interpretation of

pronouns and reflexives: Structural and semantic information. In E. Elfner & M. Walkow

(Eds.), Proceedings of the North East Linguistic Society, 37, GLSA Publications.



Kaiser, E., & Trueswell, J. C. (2004). The role of discourse context in the processing of a flexible

word-order language. Cognition, 94(2), 113�147.Kaiser, E., & Trueswell, J. C. (in press). Investigating the interpretation of pronouns and

demonstratives in Finnish: Going beyond salience. In E. Gibson & N. Pearlmutter (Eds.),

The processing and acquisition of reference. Cambridge, MA: MIT Press.

Kaiser, E., & Vihman, V. (2006). On the referential properties of Estonian pronouns and

demonstratives. Presented at the 22nd Scandinavian Conference of Linguistics, to appear in the

Proceedings of the 22nd Scandinavian Conference on Linguistics.

Kaiser, E., & Vihman, V. (2008). Reference resolution in Estonian: What guides the interpretation of

personal pronouns and demonstratives? Manuscript in preparation. University of Southern

California and University of Tartu.

Kalliokoski, J. (1991). Emphathy as motivation for style shifting in narrative. In J. Verschueren

(Ed.), Levels of linguistic adaptation. Selected papers of the International Pragmatics Conference

Vol II (pp. 147�161). Philadelphia/Amsterdam: John Benjamins.

Kameyama, M. (1996). Indefeasible semantics and defeasible pragmatics. In M. Kanazawa, C.

Pinon, & H. de Swart (Eds.), Quantifiers, deduction and context (pp. 111�138). CSLI

Publications: Stanford.

Kameyama, M. (1999). Stressed and unstressed pronouns: Complementary preferences. In P. Bosch

& R. van der Sandt (Eds.), Focus: Linguistic, cognitive, and computational perspectives (pp. 306�321). Cambridge, UK: Cambridge University Press.

Kiss, K. (1995). Discourse Configurational Languages. New York: Oxford University Press.

Laitinen, L. (1992). Valttamattomyys ja persoona. Suomen murteiden nesessiivisten rakenteiden

semantiikkaa ja kielioppia. Helsinki, Finland: Suomalaisen Kirjallisuuden Seura.

Lappin, S., & Leass, H. (1994). An algorithm for pronominal anaphora resolution. Computational

Linguistics, 20(4), 535�561.Levy, E. (1982). Toward an objective definition of discourse topic. In K. Truite, R. Schneider, & R.

Chametzky (Eds.), Papers from the eighteenth regional meeting of the Chicago Linguistic Society

(pp. 259�304). Chicago, IL.Matin, E., Shao, K. C., & Boff, K. R. (1993). Saccadic overhead: Information processing time with

and without saccades. Perception and Psychophysics, 53(4), 372�380.Prasad, R., & Strube, M. (2000). Discourse salience and pronoun resolution in Hindi. In A.

Williams, & E. Kaiser (Eds.), U. Penn Working Papers in Linguistics, 6.3, 189�208.Prince, E. F. (1992). The ZPG letter: subjects, definiteness and information status. In S. Thompson

& W. Mann (Eds.), Discourse description: diverse analyses of a fund raising text (pp. 295�325).Philadelphia/Amsterdam: John Benjamins.

Rambow, O. (1993). Pragmatic aspects of scrambling and topicalization in German. Paper presented

at the Workshop on Naturally-Occurring Discourse, Institute for Research in Cognitive Science,

University of Pennsylvania, PA.

Saarimaa, E. A. (1949). Kielemme kaytanto. Pronominivirheista. Virittaja, 49, 250�257.Saslow, M. G. (1967). Latency for saccadic eye movement. Journal of the Optical Society of

America, 57, 1030�1033.Seppanen, E.-V. (1998). Lasnaolon pronominit. Tama, tuo, se ja han viittaamassa keskustelun

osallistujaan. Helsinki, Finland: Suomalaisen Kirjallisuuden Seura.

Shapiro, L. P., & Hestvik, A. (1995). On-line comprehension of VP-ellipsis: Syntactic reconstruc-

tion and semantic influence. Journal of Psycholinguistic Research, 24(6), 517�532.Shapiro, L. P., Hestvik, A., Lesan, L., & Garcia, A. R. (2003). Charting the time-course of VP-

ellipsis sentence comprehension: Evidence for an initial and independent structural analysis.

Journal of Memory and Language, 49(1), 1�19.Snedeker, J., Thorpe, K., & Trueswell, J. (2001). On choosing the parse with the scene: The role of

visual context and verb bias in ambiguity resolution. Proceedings of the 23rd Annual Conference

of the Cognitive Science Society, pp. 964�969.



Snedeker, J., & Trueswell, J. C. (2004). The developing constraints on parsing decisions: the role of

lexical-biases and referential scenes in child and adult sentence processing. Cognitive

Psychology, 49, 238�299.Strube, M., & Hahn, U. (1996). Functional centering. Proceedings of ACL ‘96, 270�277.Strube, M., & Hahn, U. (1999). Functional centering: grounding referential coherence in

information structure. Computational Linguistics, 25, 3.

Sulkala, H., & Karjalainen, M. (1992). Finnish. London: Routledge.

Tabor, W., Galantucci, B., & Richardson, D. (2004). Effects of merely local syntactic coherence on

sentence processing. Journal of Memory and Language, 50(4), 355�370.Tanenhaus, M. K., Spivey-Knowlton, M. K., Eberhard, K. M., & Sedivy, J. E. (1995). Integration

of visual and linguistic information in spoken language comprehension. Science, 268, 632�634.Tanenhaus, M., & Trueswell, J. (2006). Eye movements and spoken language comprehension. In M.

Traxler & M. A. Gernsbacher (Eds.), Handbook of Psycholinguistics (2nd ed., pp. 863�900).Amsterdam: Academic Press.

Turan, U. D. (1995). Null vs. Overt Subjects in Turkish Discourse: A Centering Analysis. Ph.D.

dissertation, University of Pennsylvania, Philadelphia, PA.

Turan, U. D. (1998). Ranking forward-looking centers in Turkish. In M. A. Walker, A. K. Joshi, &

E. F. Prince (Eds.), Centering theory in discourse (pp. 136�160). Oxford: Oxford University

Press.

Vallduvı, E. (1990). The Informational Component. Ph.D. dissertation, University of Pennsylvania,

Philadelphia, PA.

Van Dijk, T. A. & Kintsch, W. (1983). Strategies of discourse comprehension. New York: Academic

Press.

Varteva, A. (1998). Pronominit han ja tama tekstissa. Virittaja, 2, 202�223.Valimaa-Blum, R. 1988. Finnish existential clauses: their syntax, pragmatics and intonation. Ph.D.

dissertation, Ohio State University.

Vilkuna, M. (1989). Free word order in Finnish. Helsinki, Finland: Suomalaisen Kirjallisuuden

Seura.

Vilkuna, M. (1995). Discourse configurationality in Finnish. In K. Kiss (Ed.), Discourse

configurational languages (pp. 244�268). New York: Oxford University Press.

Vilppula, M. (1989). Havaintoja han- ja he-pronominien kaytosta suomen murteissa. Virittaja, 93,

389�399.



Sample continuations from Experiment 1 (sentence completion task)

(a) SVO/Han condition with subject-continuation

. . .Kokki toni jonon hanilla seisovaa leipuria. Han oli vihainen leipurille joka yritti etuilla

kassalle

. . . ‘The cook-SUBJ pushed a baker-OBJ standing at the back of the line. S/he was angry at the

baker who was trying to cut in line.’

(b) OVS/Han condition with subject-continuation

. . .Rehtoria moitti silmalasipainen kirjastonhoitaja. Han halusi rehtorin heti palauttavan

myohassa olevan kirjan.

. . . ‘The principal-OBJ criticised a librarian-SUBJ with glasses. S/he wanted the principal to

return the overdue book immediately.’

(c) SVO/Tama condition with object-continuation

. . . Sairaanhoitaja onnitteli ohikavelevaa laakaria. Tama oli suorittanut onnistuneesti vaikean

leikkauksen.

. . . ‘The nurse-SUBJ congratulated a doctor-OBJ who was walking by. This had successfully

completed a difficult operation.’

(d) OVS/Tama with subject-continuation

. . .Rehtoria moitti silmalasipainen kirjastonhoitaja. Tama oli tuohtunut rehtorin myo-

hastyneesta kirjalainasta.

. . . ‘The principal-OBJ criticised a librarian-SUBJ with glasses. This was aggravated by the

principal’s overdue book loan.’

(e) Demonstrative use of tama:

. . . Pelle pilkkasi hakin vieressa harjoittelevaa jonglooria. Tama oli Leenasta varsin huvittavaa.

. . . ‘The clown-SUBJ made fun of the juggler-OBJ practicing next to the [animal] cage. This was

rather amusing, in Leena’s opinion.’

APPENDIX A



Documents

Interpreting pronouns and demonstratives in Finnish ...emkaiser/kaiser-trueswell2008-LCP-w.pdf · Interpreting pronouns and demonstratives in Finnish: Evidence for a form-speciﬁc